Graceful shutdown for Kubernetes

[alexellis]

There appears to be an issue with graceful shutdowns. This occurs when Kubernetes sends a SIGTERM to our function's Pod.

We have two processes:

• of-watchdog
• Go handler / HTTP server

The of-watchdog catches the signal and waits for wait_timeout env-var, but the child process appears to die immediately

We need to copy the code / approach of the of-watchdog into the Go handler for all templates and test.

Task

• Show the issue by adding a SIGTERM handler and echo message in the entry point for one of the templates
• Deploy in a container
• Scale to --replicas=0
• Look at the logs

Now fix the code and repeat

You should see that both processes wait at least wait_timeout before exiting.